Investigating the Performance of Deep Reinforcement Learning-Based MPPT Algorithm under Partial Shading Condition

For renewable energy systems to operate as efficiently and as effectively as possible, maximum power point tracking (MPPT) controllers are essential. They make it possible to precisely and dynamically track the peak output of solar panels or wind turbines, ensuring that the system will be stable and reliable even in the face of changing environmental factors. Recently, more robust algorithms based on deep reinforcement learning (DRL) have been proposed. These DRL-based algorithms optimize the local and global maximum power point (MPP) using deep Q-learning and deep deterministic policy gradient (DDPG). In this study, MATLAB models of a DRL-based MPPT algorithm were developed, tested, and compared to simulation based on two established MPPT algorithms – the Particle Swarm Optimization (PSO), and the Perturb and Observe (P&O). The simulations were conducted under various conditions, including standard test conditions (STC), and partial shading conditions (PSC). Simulation results demonstrate that at STC, both the DRL-based MPPT and PSO algorithm tracks the steady-state power at 0.02 seconds, outperforming the traditional P&O technique of 0.08 seconds. However, the PSO algorithm manages to track 1.18% more power than DRL MPPT at PSC. Despite the limitations of training the DRL, it shows a promising method for addressing MPPT issues under PSC.